The Roman Galleries of Lisbon
Three dimensional reconstruction of a 1st century cryptoporticus.
Rua da Prata, Lisboa, Portugal
Date: 27th September 2007
A bit of history ...
The galleries found underneath Rua da Prata in downtown Lisbon are estimated to have been built in the 1st century a.C., during the peak of the Roman Empire. The Romans had occupied the western coast of the Iberian Peninsula since the 2nd century b.C., conquering Olissipo (Lisbon) during the campaigns of the First Punic War. Prior to the conquest of the Romans, the port on the Tagus estuary was already an important commercial port where Carthaginians traded with the local population, of celtic origins, trading metals and agricultural goods for manufactured products. The Romans began the fortification of the oppidum at Olissipo at the end of the 2nd century b.C. Between 40 and 30 b.C. Olissipo was granted Municipium Civium Romanorum status which conferred Roman Citizenship to all free born men. A time of peace and significant prosperity endured until the 3rd century A.D.. During the beginning of the Roman Empire, municipalities asserted themselves by erecting structures for public functions (forums, theatres, markets and temples): in Olissipo, traces of these buildings can still be found today. One example are the ruins of the Roman Theatre located on the same hill as the castle that overlooks modern day Lisbon. In those days, the majority of public life occurred by the side of the river. According to experts, the galleries underneath Rua da Prata were constructed to structurally stabilize the soil near the edges of the river which is normally very humid due the flow of groundwater to river from the surrounding hills. The underground construction of the cryptoporticus was necessary to provide a solid base for the construction of important buildings on this site. The considerable resources required to construct this subterranean cryptoporticus suggest that the buildings it supported where of significant importance such as the Lisbon forum or edifices dedicated to port and commercial activities.
The galleries consist of parallel corridors 3 m in height by 2 to 3 m wide. The corridors have vertical walls and circular ceilings.The principal corridors are intersected perpendicularly by narrower corridors. Some galleries have a maximum height between 1.2 and 1.5 m. Some galleries provide access to small, narrow chambers which could have been used as storage areas. The "cryptoporticus" designation refers to the hiding nature of this construction that was used as foundation to the ground buildings. Over the centuries, the roman structures on the surface crumbled or were replaced by new buildings. However, newer buildings were continuously constructed on the pre-existing roman foundations. The local inhabitants took advantage of the groundwater collected in the galleries by making boreholes above the galleries (see chimney on the right side of the image above). One of these wells came to be known as the "Poço das Águas Santas" located on the corner of Rua da Prata and Rua de São Julião.
The area encompassing the galleries (now downtown Lisbon) was completely destroyed in the earthquake of 1755. Marquês de Pombal (Prime Minister) and Manual da Maia (architect) were responsible for the reconstruction of the entire city, creating in downtown Lisbon the grid-like neighbourhood known today as the "Baixa Pombalina". The roman galleries were rediscovered during the reconstruction of the downtown area of the city. Though most buildings in the "Baixa Pombalina" (downtown) are supported by pine stakes submerged in the soft and wet soil to provide structural stability, the buildings built on top of the galleries are supported by the pre-existing roman cryptoporticus which survived the earthquake unscathed. An inscribed stone was found during early archaeological investigations of the galleries dedicated to Aesculpius, the Roman god of healing. The inscription stone and the crystalline waters found on site, led to the hypothesis that the galleries may have been used as a hot water spa. Today, this hypothesis appears completely unfounded.
In the middle of the 19th century, a storm drain and sewage project was initiated to collect all the water running through the "Baixa Pombalina". In 1859, a topographical survey of the roman galleries was completed as part of this project. Unfortunately, the storm drains intersect the galleries along the roads "Rua da Prataria (Rua da Prata") and "Rua dos Retrozeiros (Rua da Conceição)" making the largest part of the galleries inaccessible since the location of all but two entry point from the surface have been lost.
Blueprint view of the three dimensional model of the galleries. The model shown does contain all chambers and galleries. The principal axis of the galleries is perpendicular to the margin of the river Tagus which, if represented in the image, would be located beneath the image (approximately directly South of the galleries).
Detail of the transverse chambers. On the right, the chimney-like structure is in reality an opening for well.
A city worker washes the entrance to the galleries in the middle of the road between two tramways on "Rua da Conceição"
The number 28 tram passes right next to the entrance to the galleries.
The galleries today....
The galleries continue to accumulate groundwater due to a long crack along the floor of the gallery designated as "A Galeria das Nascentes". As a result, the cryptoporticus is always partially submerged. Several years ago, "Museu da Cidade", the museum responsible for the monument operating under the "Câmara Municipal de Lisboa" (Lisbon's Municipality), decided to allow visits to the monument but only for a few days a year. Public access is only possible after pumping out all the water accumulated over the entire year and the installation of lights. Water pumps continue work until the monument is once again closed to the public due to the significant rate of water seepage. Public access is granted for only 2 to 3 days a year, drawing large crowds who wait patiently in queue for more than a hour to visit the galleries. Access to the galleries is in a very precarious location, in the middle of the road ("Rua da Conceição") next to the tramway as shown below. Since the trams cannot alter their path, visitors must enter and exit galleries with extreme caution.
The proposal for 3D reconstruction
Albatroz Engineering proposed reconstructing the a 3D model of the Roman Galleries to "Museu da Cidade". This way, not only would archaeologists possess an accurate model of the galleries but people would be able to explore the galleries, which are difficult to visit in situ, in a virtual environment.
The reconstruction involved the use of a Laser Range Scanner that measures distances to objects in the vicinity of the laser.
Data acquisition required installation of a support structure so that the laser could traverse the entire span of the galleries. Since the equipment used is designed to operate in harsh environments and is eye safe, equipment transportation, installation and data acquisition were completed while the galleries were being prepared for public access.
The accessible area of the roman galleries is comprised of six galleries of varying lengths between 12 m and 24 m and several smaller chambers.
Horizontal section of the three dimensional model. Not all galleries and chambers are shown here. The large gallery connected to the entrance is shown vertically on the left while the gallery known as the "Galeria das Nascentes" is shown vertically on the extreme right.
What is a laser scanner?
A laser range scanner is a sweep sensor composed of three parts:
- A visible or invisible light source. The emitter creates a concentrated ray of light.
- Component that detects the portion of the ray of light that is reflected after it comes into contact with a target. The quantity of light reflected is dependant on the colour and nature of the target.
- Sweep (scan)
- Mechanical system that rotates the ray of light over an arc of a circle using a sophisticated system of mirrors.
Archaeologist, António Marques, presents the galleries to the media...
... while the Albatroz Engineering Team acquires data.
The virtual reconstruction proposed by Albatroz Engineering is based on a three dimensional model created using clouds of individual coordinates (xi, yi, zi). These types of models can be observed, rotated, navigated and manipulated in three dimensions.
Polygon surfaces can be used to unite the three dimensional coordinates thereby generating solid floors, walls and ceilings.
On the right there is a picture, taken in situ, of a section of the entrance gallery, the point cloud corresponding to this section of entrance gallery and finally, an image of the surfaces generated from the point cloud. Of interest is the water filled area of the floor and the two flagstones resting against the gallery wall on the right.
The number of points necessary to construct the surface model can be significantly reduced, especially in cases where the surfaces are regular and very nearly flat. Thus, the representation of the galleries can be simplified using larger surfaces which lighten the graphical computational load without sacrificing the general geometry of the model. As an example, compare the model of the entrance gallery using 155662 points with the surface model using only 8474 points (or approximately 5%).
Fundamentals of 3D model generation
Construction of the 3D model is based on subsequent 2D laser sweeps along each of the galleries. The animation in the following figure illustrates the process. On the left, the laser is shown as it traverses the galleries and on the right, the raw laser data.
To create the 3D model either the laser is moved within a specific environment (such as a gallery), or the scenario is moved while the laser is held in a fixed position. Since archaeological sites have dimensions several orders of magnitude greater than the laser, it is far easier to move the laser through site. One of the easiest and cost effective ways to traverse the laser sensor through the site is to use an autonomous or semi-autonomous moving platform. The key to a successful model is register and synchronize the position of the sensor over time in the reference frame of the sweeping plane.
The image on the left illustrates how a laser sensor collects data as it moves along the same section of the gallery as depicted in the figure above. The laser performs a 360º sweep, represented in red, as it is traverses the gallery along its major axis. With each sweep the laser measures between 500 and 1500 points and each point is represented by the distance of the laser to the nearest obstacle in each direction. The data collected in a sample sweep shown on the right in the figure below. The laser was located approximately in centre of the gallery, circa 0.7 m above the ground. For comparison, the principal gallery (entrance gallery) is approximately 2.3 m in height while the perpendicular gallery on the left is approximately 1.4 m in height.
The laser measures the surrounding volumetric free space while the user "sees" the surfaces that the limit this free space. Analysis of the animation on the right reveals that the free space in the gallery increases as the sensor approaches a intersecting gallery or a "chimney" that is really an old water well inside a building on Rua da Conceição. Also, light fixtures are detected on the right hand side of the gallery by the abrupt change in the smooth curvature of ceiling.
One of the drawbacks of collecting data in this manner is the occlusion of all surfaces that are behind other objects present on site. This phenomenon can be observed in the shadows present in the previous figure or behind a lighting fixture in the figure on the right:
The experimental structure used for data acquisition is illustrated in the next figure below. A small push cart that carries sensors, batteries, and a computer glides along an aluminium rectangular monorail. The computer records and provides data acquisition feedback and imaging in real time.
The data is acquired, recorded and visualized in real time. The GUI permits visualization of a partial 3D reconstruction of several different galleries as shown in the figure on the right.
1 There are laser reconstruction applications where the laser remains fixed and it is the target that is moving e.g. rotating a small statue in the laser field of view.
Determining the volume of free space surrounding the laser.
Acquisition system monorail car used to traverse the galleries along the longitudinal axis.
Acquisition application with 2D visualization in real time.
How does a laser scanner work?
A laser scanner measures distances based on a principle called "time of flight":
- The emitter sends out a pulse of light.
- The target reflects a fraction of the incident light depending on the colour, texture, and angle of incidence. The rest of the light is absorbed or dispersed around the point of incidence.
- The receiver detects the reflected light and calculates the distance to the object by dividing the time of flight by the speed of light.
- The sweeping system rotates in order to point the ray to a different location very near the previous point of incidence restarting the entire process.
Partial 3D Models
Once the 2D sweeps of the entire site have been acquired, as illustrated in the preview figures above, the data must be assembled so that the 3D model can be generated. Assembling the data requires measurements of the change in position of the laser sensor as it traverses the site. This data is acquired using auxiliary sensors. To minimize measurement errors, the laser should be moved along rectilinear paths. If needed, as in the case of the Roman Galleries, the reconstruction required superimposing several rectilinear perpendicular paths. Other paths may be better suited to the site that is being reconstructed. The associated position errors, however, will have a tendency to increase.
The 3D model of a section of the entrance gallery is shown in the figure in the right.
- The longitudinal resolution, or the distance between successive sweeps, may vary according to the reconstruction objectives and nature of the site. The distance between successive sweeps is dependant on the speed of the moving platform. In this case, the distance between successive scans was approximately 5 cm and is noticeable in the the figure above.
- The sweeping resolution, or distance between adjacent points in the same sweep, is an intrinsic feature of the sensor used. That is, the sweeping resolution varies with types and brands of laser sensors. In each sweep, the light ray outlines a circular plane in which the pulses of light are equally spaced. The distance between each adjacent point varies linearly with the dimensions of the site. In the case of the galleries, this distance is approximately 1 cm. At this resolution, the points appear to be touching as in the figure above.
- The distribution points in 3D space is heterogeneous and the point density is dependent on the axis of movement.
- Only when distances to the surface increase and the surface is not perpendicular to the laser beam is the sweep discontinuous. This can be seen with the "chimney" in the figure above.
- The systematic lack of points in the centre of the gallery corresponds to the monorail.
- The stone reliefs are visible on the inside of the gallery.
- The areas of the ground not covered by stone but by water are not detected. Given the depth of the water on the ground, the light nearly entirely absorbed reflecting too weak a signal for the laser to make a measurement in those specific directions.
This 3D model can be represented and explored in 3D using appropriate software tools. One way to visualize the space is through VRML - Virtual Reality Modelling Language. An example is provided below.
The model below was constructed from the data collected after the number of points had been reduced and surfaces generated that connect adjacent vertices.
This surface model, clearly illustrates the differences between the projected reality and the reconstructed reality. While projected models are formed by regular surfaces with rectangular edges and planar surfaces, the reconstructed model incorporates all the small irregularities of the surface of the gallery (also due, in a very small part, to measurement error), providing a more accurate model of the site.
The data shown in VRML can be imported into Computer Aided Design (CAD) software for manipulation, analysis and optimization by architects, civil engineers and archaeologists. With specialized software accurate architectural models can be created combining already existent buildings with virtual, to-be-constructed, elements or even structures that existed in the past and have long disappeared.
Click on the image to explore the 3D model. [model 7Mb; VRML plugin required]
3D model of a section of the entrance gallery with complete data set of points for this section. A different view of this section is presented above.
Characterization of a laser scanner
A sweeping laser sensor is characterized by six main parameters that determine the applications to which it is suited.
The first three parameters relate to internal performance while the last three parameters characterize the environment in which the sensor operates.
- Range is the maximum distance from the laser an object can be detected. Most lasers also have a minimum measurement distance.
- Field of View (FOV)
- Field of view determines the angular span of the beam of light. Albatroz Engineering possesses a laser with 360º FOV and one with a 60º FOV but of significantly higher resolution for tasks requiring greater precision.
- Acquisition Rate
- Acquisition rate refers to the number measurements made in a specified interval of time. This characteristic is crucial in determining which laser is best suited for each application. For example, a 3D model of a statue can be made with a laser with very low acquisition rate. To detect automobiles at a tollgate, however, requires a laser with a very high acquisition rate.
- The class determines the inherent risk of exposure to the laser beam. All lasers used by Albatroz Engineering are Class 1 lasers which are eye safe for those in the vicinity of the laser. Common Class 1 lasers are used in CD/DVD players.
In addition to the limited beam strength, the lasers used have rotating internal mirrors or rotating heads which change the direction of the laser beam very rapidly reducing the accumulated radiation at each illuminated point.
- IP (Ingress Protection) Index
- The IP index provides a measure of protection against adverse environmental conditions. External protection means should never be used over the mirrors of the sensor for these may hinder measurement efficacy.
- Dimensions, Mass and Power
- Physical dimensions, mass and power requirements define the suitability of the laser for different applications. A laser scanner is a relatively heavy device with a mass of ranging from 5 kg to 50 kg while a video camera has a mass of approximately 0.5 kg. Power requirements are also important. A typical laser scanner requires anywhere between 25 W and 250 W. For comparison, a video camera requires less than 5 W.
Integrated 3D Model
In order to integrate every part (gallery) of the model, all date in local coordinates (xi, y i, zi) must be converted to a common general gallery reference frame. In the roman galleries, nine data sets were collected of the six galleries.
Data integration required careful analysis of overlapping data corresponding to the same sections of the galleries acquired during different data acquisition procedures. The following figure shows the data acquisition sequence beginning with the entrance gallery. In the final model (floorplan above), galleries 5 and 6 were not included due to relative humidity very near saturation and the presence of water pumping equipment. These two factors did not permit a model of these galleries of sufficient quality to be included in the general model.
An alternative method of visualization is to create clouds of animated points, observed from different points of view, as ilustrated in the videos em baixo.
These models may further enriched by "colouring" each point in the point cloud according to the different properties of their associated objects in the sensor surroundings. For example: the level of reflectivity (amount of energy reflected after light beam encounters an object.), temperature of radiating surfaces on site or even just colours visible with the human eye. It is also possible to create lighting effects which make the model more realistic.
Finally, an animation is provided of a model integrating four of the six galleries. As mentioned previously, galleries 5 and 6 are not included in the model due to adverse conditions found in those chambers during acquisition. In addition to galleries 5 and 6, the integrated model also lacks:
- one chamber on the south side of gallery 1 (entrance gallery)
- one small chamber on the north side of gallery 3 (parallel to gallery 1 - "Galeria das Nascentes")
- tank located between galleries 5 and 6
A point cloud can be used for more than simply data visualization since each point is referenced in 3-D space. Thus, the point cloud can also be used to calculate lengths, areas and volumes.
A laser scanner can be used in stone quarries, mining piles, landfills and other large scale structures and areas to determine volumes or volumes that have been removed.
A laser scanner can also measure safety distances and areas in construction projects for quality control.
At the beginning of archaeological excavations, a laser scanner can be used to rapidly generate a 3D model of the site (ground, vegetation and surrounding structures) so that archaeologists can estimate the locations of paths and support installations. During the dig, a quick model is useful for assessing progress and overall geometry of the excavation. Once completed, 3D models of artefacts and ruins combined with still and video images can create stunning multimedia content which help to divulge the site.
In reconstruction architecture, existing structures can be modelled and act as the starting point for the project. Architectural elements can then be superimposed or removed easily as required.
By default, all images are propriety of Albatroz Engineering except logos that belong to the respective brands and institutions. Albatroz Engineering would like to thank those that made their pictures and images available to Albatroz Engineering. These are identified throughout the text.
If you are looking for a VRML client in order to visualize 3D VRML models try Cortona VRML.